> For the complete documentation index, see [llms.txt](https://fieldworker.gitbook.io/fieldworker-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://fieldworker.gitbook.io/fieldworker-docs/overview/what-next.md).

# What next?

LLMs and agentic AI

The convergence of Large Language Models (LLMs) and agentic AI is creating a massive paradigm shift in Home and Community-Based Services (HCBS) for seniors and individuals with Intellectual and Developmental Disabilities (IDD). The true impact isn’t about replacing the human element of care; it is about shifting Direct Support Professionals (DSPs) and case managers from being administrative data-entry clerks back to being caregivers.

### The Reality

The social care sector is facing a severe dual pressure: compounding caregiver shortages alongside an increasingly aggressive regulatory environment.

The federal government is deploying its own AI tools (such as the HHS AERO initiative) to audit multi-year Medicaid claims history for fraud and waste at scale. Concurrently, strict state-level legislation requires absolute transparency and mandates a human-in-the-loop for any AI-assisted claim or clinical note. For platforms managing care coordination and service integrity (like Fieldworker), the mandate is clear: move from a *passive system of record* to an *active system of intelligence and defense*.

### Strategic Roadmap

To position technology as an indispensable shield and partner for providers and other care coordination agencies, the Fieldworker management envisions enhanced capabilities in three distinct areas.

#### **Ambient Capture & Administrative Relief**

Focus heavily on the direct caregiver’s immediate pain point: documentation burnout.

* Voice-to-ISP Mapping: Implement secure, device-agnostic ambient listening or structured dictation. A caregiver should be able to speak naturally about their shift ("Helped John with his morning routine, he took his meds but seemed a bit more lethargic than usual...").
* Automated Structuring: The backend LLM must immediately clean, translate (supporting non-native English speaking DSPs), and map that narrative directly to Individualized Service Plan (ISP) goals and Electronic Visit Verification (EVV) milestones.

#### **Agentic Service Integrity & Cross-entity Synthesis**

Move up the value chain to protect the agency's bottom line by preventing claim denials before they happen.

* Pre-Claim Compliance Auditing: Develop an LLM-driven "compliance copilot" that reads daily activities and cross-references them across multiple data pools—historical care notes, scheduled hours, and state-specific Medicaid rules.
* Discrepancy Flagging: If an automation layer detects a gap (e.g., a note mentions a behavioral intervention, but the required behavioral tab or specific form wasn't completed), the system flags it for a billing professional *before* submission, protecting the agency from retroactive downcoding.

#### **Predictive Coordination & The Circle of Care**

Leverage aggregated data to move from reactive compliance to proactive, predictive care management.

* Trend & Escalation Detection: Deploy smaller, fine-tuned models to analyze behavioral notes across weeks. If an IDD patient or senior shows a slow, multi-week decline in mobility or a subtle uptick in targeted behaviors, the system automatically surfaces a proactive alert to the case manager.
* The Enhanced Family Hub: Build out a consumer-facing layer that synthesizes complex, clinical daily shift notes into plain-language, reassuring, and secure summaries for family members and guardians, drastically reducing manual communication overhead for agency directors.

> Critical Guardrail for Fieldworker:
>
> 2026 state compliance frameworks are strictly penalizing autonomous AI decision-making in healthcare and social services. Every AI feature built into the workflow explicitly present outputs as *drafts* or *recommendations*, requiring a licensed professional or billing coordinator to review (Human in the loop) and sign off. The software must maintain an audit-ready log showing exactly where the human modified or approved the AI's work.